Anti-jerk devices are presently available for a fuel injection internal combustion engine. The injection time in such a well-known injection system is determined by the discharge time of a capacitor whose charging depends on the rpm and whose discharging depends on the load. In order to implement the anti-jerk mechanism of such a known device, the charging process proceeds in two phases in which, during a first phase, the charging occurs relatively fast up to a certain signal level and then it levels off. In this way, sudden changes in the rpm do not cause large differences in the amount of fuel injected and torque fluctuations are thus prevented. It has become evident that this well-known anti-jerk device does not work satisfactorily since, in the actual anti-jerk device, only an rmp signal is used. Therefore optimum driving comfort is not achieved. In regard to optimum acceleration of the internal combustion engine, the limitation of the alteration should also be dependent on the existing load.